You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@horn.apache.org by "Edward J. Yoon (JIRA)" <ji...@apache.org> on 2016/05/16 02:38:12 UTC
[jira] [Commented] (HORN-23) Add softmax function
[ https://issues.apache.org/jira/browse/HORN-23?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15284082#comment-15284082 ]
Edward J. Yoon commented on HORN-23:
------------------------------------
Basically output needs to be divided by the sum of exponential of the whole output vector. So, it's impossible to define clearly within two forward/backward methods of current neuron-centric programming model.
There're two options. 1) calculate sum as below
{code}
if (squashingFunction.getFunctionName().equalsIgnoreCase(
SoftMax.class.getSimpleName())) {
// divide by the sum of exponential of the whole vector
vec = vec.divide(vec.sum());
// LOG.info("softmaxed output: " + Arrays.toString(vec.toArray()));
}
{code}
2) add more APIs for handling intermediate output.
> Add softmax function
> --------------------
>
> Key: HORN-23
> URL: https://issues.apache.org/jira/browse/HORN-23
> Project: Apache Horn
> Issue Type: New Feature
> Reporter: Edward J. Yoon
>
> https://en.wikipedia.org/wiki/Softmax_function
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)